Conceptual questions about GPU acceleration

From: Andrew DeYoung (adeyoung_at_andrew.cmu.edu)
Date: Sat Apr 27 2013 - 11:46:49 CDT

Hi NAMD experts,

I am a graduate student and have used MD codes on a Linux CPU cluster for
the past 2 years. However, I am a complete beginner when it comes to GPUs
and GPU acceleration of MD.

This is a very basic question, and I am a little embarassed to be asking it,
but I will ask it anyway. Do you know of any good review articles that
would help me to understand what a GPU is and how it can be used to improve
the performance of MD codes such as NAMD?

Even after reading articles on Wikipedia and Nvidia websites, I am still
confused as to what a GPU is (I have no background in computer
architecture). I understand that GPUs were originally designed for graphics
computations, but in the last 10 or 15 years, they have been used for
high-performance applications because they are efficient at floating point
arithmetic. I think that GPUs must be used in tandem with at least one CPU.

I have read the following excellent review article by Stone et al. from
2010:

Stone, John E.; Hardy, David J.; Ufimtsev, Ivan S.; Schulten, Klaus. Journal
of Molecular Graphics and Modelling 2010, 29, 116-125. "GPU-accelerated
molecular modeling comes of age."

But, I feel that it would help me to read an even more basic tutorial/review
of GPUs and their application to, for example, MD. Do you know of such
tutorials, reviews, or websites? Thank you for considering this somewhat
off-topic question!

Andrew DeYoung
Carnegie Mellon University

This archive was generated by hypermail 2.1.6 : Tue Dec 31 2013 - 23:23:10 CST